1. Field of the Invention
The present invention relates to pattern recognition method and device for position matching between a flat substance having a regularly repeating pattern on the surface thereof such as cloth, semiconductor wafer, wood board, metal plate, leather and paper, and a landscape photograph in which buildings and natural matters regularly appear, and a reference figure, and, in particular, relates to a pattern recognition method and device for position matching between a cloth with a pattern which is an object for cutting and a predetermined reference cutting pattern via an image processing, and relates to an automatic pattern match cutting device which performs cutting of the cloth with a pattern according to the predetermined reference cutting pattern.
2. Description of Related Art
Heretofore, in the case of pattern match cutting of a cloth with a pattern, the cloth is cut manually after a paper pattern is put on the cloth. Because such manual cutting is inferior in efficiency to the automated cutting of a cloth without a pattern, a demand for automation of pattern matching has been strong. Cutting devices directed to the automation of pattern matching are described, for example, in JP-A-1-250465(1989), JP-B-1-33587(1989) and the like.
In JP-A-1-250465(1989) a system for performing pattern matching while moving the contour of parts of cutting pattern through an operator after superposing both the pattern of a cloth from a camera and the contour of parts of cutting pattern on a display is disclosed. According to the system, a direct cutting can be made without use of a paper pattern.
In JP-B-1-33587(1989), which corresponds to U.S. Pat. No. 4,853,866, a full automatic pattern match cutting device is realized by performing pattern recognition of a cloth having a pattern through an image processor. This is a system in which the effect can be expected in the case of pattern match cutting of a cloth having a clear pattern. Further, a method in which the operator performs pattern matching manually by using an image on a monitor and a digitizer when automatic pattern matching is impossible is disclosed therein.
In JP-A-4-240261(1992), which corresponds to U.S. patent application Ser. No. 07/825,193 filed on Jan. 24, 1992 and Spanish patent application No. P 9200142 filed on Jan. 23, 1992, an automatic pattern match cutting device which performs an automatic pattern recognition via image analysis as well as a measure permitting intervention of an operator for dealing with a subtle pattern when there appears a pattern matching point which is difficult for the automatic pattern recognition. The device uses a method of locally matching X axis or Y axis concentration accumulated projection histogram for the pattern recognition, and a key position on a similar configuration portion is determined via a matching process between a characteristic portion of a sampled histogram during a teaching operation and a histogram generated based on an image of a recognition object.
Further, in JP-A-63-98070(1988), a method in which small regions of two dimensional data are sequentially selected and their correlation is processed in high speed is disclosed.
Other than the above conventional methods, a binary template matching method and a concentration correlating pattern matching method are well known.
However, all of these conventional pattern recogition methods are ones which compare a pattern form in a local area with a reference data and are for recognizing a pattern on a surface of a hard substance.
There are several conventional methods which are thought effective for recognizing pattern forms as indicated above, however, when these methods are actually applied to a cloth, their recognition rates are far more inferior than those expected. The reasons why a recognition of a pattern form on a cloth is difficult are due to the following features inherent to a cloth.
(1) Meshes and weaves inherent to the surface of a cloth operate as noises during the image recognition thereof. Moreover, when a smoothing process and the like for erasing signals relating to the meshes and weaves are performed, the information relating to the pattern form itself can be lost. PA0 (2) A cloth is likely to be subjected to expansion and contraction and deformation due to external forces so that it is difficult to expect always a coincidence of pattern forms during pattern matching operation between pattern forms of a standard template used in such as the conventional binary template matching method and the conventional concentration pattern matching method, and of an object cloth. PA0 (3) Kinds of patterns which can be used for pattern matching are limited to checked patterns and striped patterns among many pattern forms of cloths. The task of pattern recognition for the checked patterns and the striped patterns is to determined a starting point of the pattern form which is repeated with a predetermined pitch. However, with the conventional pattern recognition methods which are performed via comparison of a local configuration, it was difficult to fully achieve the task and it will be difficult to achieve an accurate pattern recognition, other than a pattern recognition method which permits selective use of all of the characteristic quantities of a concentration accumulated projection histogram generated from a pattern form over one pitch such as number of crests, position thereof, width thereof, height thereof, area thereof, height difference of crests, number of bottoms, position thereof, width thereof, depth thereof, area thereof, depth difference of bottoms, position and width of flat portions and position and width of crowded crests depending upon necessity of individual pattern recognitions. PA0 (4) In particular, many pattern forms on cloths for men's wear are subtle ones which are sometimes difficult to recognize as having a pattern form even with human eyes if the inspection is performed partially. In such instance, it is thought that there is no proper information other than a concentration acumulated projection histogram taken along the X axis and Y axis on an input screen image of a cloth. PA0 (1) Since patterns applicable for pattern matching are limited to checked patterns and striped patterns, the vertical lines and the lateral lines in the respective pattern forms constitute their fundamental elements, therefore, X axis and Y axis concentration accumulated projection histograms are prepared and the prepared X axis and Y axis histograms are independently analyzed to separately obtain characteristic quantities for the vertical pattern forms and the lateral pattern forms. PA0 (2) All of the characteristic quantities in the generated histograms in connection with the vertical lines and lateral lines within one pitch are noted and one or more optimum characteristic quantities are selected manually or automatically for the pattern recognition for individual pattern forms of cloths. PA0 (3) The histograms prepared based upon an input screen image according to the above method show a characteristic of which center portion rises due to a shading caused by influences of an optical system and an electronic system in a device used for the image processing. Accordingly, these histograms are corrected in such a manner that the center portion and both end portions of the histograms are maintained at same levels. An advantage of this correction method resides in a fact that a desirable correction effect is obtained regardless of lightness variation of objects for pattern recognition. PA0 (4) In case that vertical lines and/or lateral lines constituting a pattern form are inclined, an error caused by the inclination is corrected to maintain a predetermined accuracy for respective characteristic quantities in histograms. PA0 (5) A pattern is recognized by selecting characteristic quantities such as relating to a remarkable crest, a plurality of crests and a bottom within one pitch in a histogram. PA0 (6) When a histogram over one pitch is divided into two parts, one shows an intense variation and the other shows a slight variation, the difference of the states is used as a characteristic quantity for the recognition. PA0 (7) When a histogram over one pitch is divided into two parts, one shows a light appearance and the other shows a dark appearance, the difference of lightness is used as a characteristic quantity for the recognition. PA0 (9) The histograms prepared based upon an input screen image according to the above further includes small noises caused by influences of an optical system and an electric system in a device used for the pattern recognition. Accordingly, these small noises are eliminated by averaging these histograms. PA0 (10) At least one of the height difference of crests, the crest area, the crest width, depth difference of bottoms, the bottom area, the bottom width and the height difference between crest and bottom in a prepared histogram are used as characteristic quantities for the recognition. PA0 (11) A second histogram is prepared which enables a pattern recognition even when a size of the pattern form slightly deviates between teaching and actual pattern recognition by allocating a predetermined band for a characteristic quantity. PA0 (12) In case when a top of a crest in a histogram is distorted and can not be used as a pattern key position during pattern key position determination, a true pattern key position can be determined by correcting the histogram.
For achieving the above tasks, different from the conventional methods which include a comparison of a local configuration, a method which makes use of all of the characteristic quantities of a generated projection histogram such as on vertical lines and lateral lines over one pitch of the pattern form for the pattern recognition is proposed, thereby a pattern recognition method, in other words, a pattern recognition algorithm which can be applied to variety of pattern forms is realized.